@annika yeah this was Grady's point - for Durable software, you either don't use AI, or you use it for very specific tasks and you verify the output. (his entire argument is that you still need knowledge of computer science to produce good software, LLMs are just a new abstraction layer to creating it)